h264: stricter reference limit enforcement.
[ffmpeg.git] / libavcodec / h264.c
index 928a96a..e0eb8e1 100644 (file)
@@ -27,6 +27,8 @@
 
 #include "libavutil/imgutils.h"
 #include "internal.h"
+#include "cabac.h"
+#include "cabac_functions.h"
 #include "dsputil.h"
 #include "avcodec.h"
 #include "mpegvideo.h"
@@ -40,8 +42,6 @@
 #include "vdpau_internal.h"
 #include "libavutil/avassert.h"
 
-#include "cabac.h"
-
 //#undef NDEBUG
 #include <assert.h>
 
@@ -56,12 +56,14 @@ static const uint8_t div6[QP_MAX_NUM+1]={
 static const enum PixelFormat hwaccel_pixfmt_list_h264_jpeg_420[] = {
     PIX_FMT_DXVA2_VLD,
     PIX_FMT_VAAPI_VLD,
+    PIX_FMT_VDA_VLD,
     PIX_FMT_YUVJ420P,
     PIX_FMT_NONE
 };
 
 /**
- * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
+ * Check if the top & left blocks are available if needed and
+ * change the dc mode so it only uses the available blocks.
  */
 int ff_h264_check_intra4x4_pred_mode(H264Context *h){
     MpegEncContext * const s = &h->s;
@@ -100,9 +102,10 @@ int ff_h264_check_intra4x4_pred_mode(H264Context *h){
 } //FIXME cleanup like ff_h264_check_intra_pred_mode
 
 /**
- * checks if the top & left blocks are available if needed & changes the dc mode so it only uses the available blocks.
+ * Check if the top & left blocks are available if needed and
+ * change the dc mode so it only uses the available blocks.
  */
-int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
+int ff_h264_check_intra_pred_mode(H264Context *h, int mode, int is_chroma){
     MpegEncContext * const s = &h->s;
     static const int8_t top [7]= {LEFT_DC_PRED8x8, 1,-1,-1};
     static const int8_t left[7]= { TOP_DC_PRED8x8,-1, 2,-1,DC_128_PRED8x8};
@@ -122,7 +125,7 @@ int ff_h264_check_intra_pred_mode(H264Context *h, int mode){
 
     if((h->left_samples_available&0x8080) != 0x8080){
         mode= left[ mode ];
-        if(h->left_samples_available&0x8080){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
+        if(is_chroma && (h->left_samples_available&0x8080)){ //mad cow disease mode, aka MBAFF + constrained_intra_pred
             mode= ALZHEIMER_DC_L0T_PRED8x8 + (!(h->left_samples_available&0x8000)) + 2*(mode == DC_128_PRED8x8);
         }
         if(mode<0){
@@ -438,11 +441,14 @@ static void chroma_dc_dct_c(DCTELEM *block){
 }
 #endif
 
-static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square, int chroma_height, int delta, int list,
-                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
-                           int src_x_offset, int src_y_offset,
-                           qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
-                           int pixel_shift, int chroma444){
+static av_always_inline void
+mc_dir_part(H264Context *h, Picture *pic, int n, int square,
+            int height, int delta, int list,
+            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+            int src_x_offset, int src_y_offset,
+            qpel_mc_func *qpix_op, h264_chroma_mc_func chroma_op,
+            int pixel_shift, int chroma_idc)
+{
     MpegEncContext * const s = &h->s;
     const int mx= h->mv_cache[list][ scan8[n] ][0] + src_x_offset*8;
     int my=       h->mv_cache[list][ scan8[n] ][1] + src_y_offset*8;
@@ -457,6 +463,7 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
     const int full_my= my>>2;
     const int pic_width  = 16*s->mb_width;
     const int pic_height = 16*s->mb_height >> MB_FIELD;
+    int ysh;
 
     if(mx&7) extra_width -= 3;
     if(my&7) extra_height -= 3;
@@ -465,7 +472,8 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
        || full_my < 0-extra_height
        || full_mx + 16/*FIXME*/ > pic_width + extra_width
        || full_my + 16/*FIXME*/ > pic_height + extra_height){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize, 16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_y - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
+                                16+5, 16+5/*FIXME*/, full_mx-2, full_my-2, pic_width, pic_height);
             src_y= s->edge_emu_buffer + (2 << pixel_shift) + 2*h->mb_linesize;
         emu=1;
     }
@@ -477,7 +485,7 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
 
     if(CONFIG_GRAY && s->flags&CODEC_FLAG_GRAY) return;
 
-    if(chroma444){
+    if(chroma_idc == 3 /* yuv444 */){
         src_cb = pic->f.data[1] + offset;
         if(emu){
             s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb - (2 << pixel_shift) - 2*h->mb_linesize, h->mb_linesize,
@@ -502,42 +510,55 @@ static inline void mc_dir_part(H264Context *h, Picture *pic, int n, int square,
         return;
     }
 
-    if(MB_FIELD){
+    ysh = 3 - (chroma_idc == 2 /* yuv422 */);
+    if(chroma_idc == 1 /* yuv420 */ && MB_FIELD){
         // chroma offset when predicting from a field of opposite parity
         my += 2 * ((s->mb_y & 1) - (pic->f.reference - 1));
         emu |= (my>>3) < 0 || (my>>3) + 8 >= (pic_height>>1);
     }
-    src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> 3) * h->mb_uvlinesize;
-    src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> 3) * h->mb_uvlinesize;
+
+    src_cb = pic->f.data[1] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
+    src_cr = pic->f.data[2] + ((mx >> 3) << pixel_shift) + (my >> ysh) * h->mb_uvlinesize;
 
     if(emu){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cb, h->mb_uvlinesize,
+                                9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
+                                pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
             src_cb= s->edge_emu_buffer;
     }
-    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, chroma_height, mx&7, my&7);
+    chroma_op(dest_cb, src_cb, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
+              mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
 
     if(emu){
-        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize, 9, 9/*FIXME*/, (mx>>3), (my>>3), pic_width>>1, pic_height>>1);
+        s->dsp.emulated_edge_mc(s->edge_emu_buffer, src_cr, h->mb_uvlinesize,
+                                9, 8 * chroma_idc + 1, (mx >> 3), (my >> ysh),
+                                pic_width >> 1, pic_height >> (chroma_idc == 1 /* yuv420 */));
             src_cr= s->edge_emu_buffer;
     }
-    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, chroma_height, mx&7, my&7);
+    chroma_op(dest_cr, src_cr, h->mb_uvlinesize, height >> (chroma_idc == 1 /* yuv420 */),
+              mx&7, (my << (chroma_idc == 2 /* yuv422 */)) &7);
 }
 
-static inline void mc_part_std(H264Context *h, int n, int square, int chroma_height, int delta,
-                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
-                           int x_offset, int y_offset,
-                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
-                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
-                           int list0, int list1, int pixel_shift, int chroma444){
+static av_always_inline void
+mc_part_std(H264Context *h, int n, int square, int height, int delta,
+            uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+            int x_offset, int y_offset,
+            qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
+            qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
+            int list0, int list1, int pixel_shift, int chroma_idc)
+{
     MpegEncContext * const s = &h->s;
     qpel_mc_func *qpix_op=  qpix_put;
     h264_chroma_mc_func chroma_op= chroma_put;
 
     dest_y  += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
-    if(chroma444){
+    if (chroma_idc == 3 /* yuv444 */) {
         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
-    }else{
+    } else if (chroma_idc == 2 /* yuv422 */) {
+        dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
+        dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
+    } else /* yuv420 */ {
         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
     }
@@ -546,9 +567,9 @@ static inline void mc_part_std(H264Context *h, int n, int square, int chroma_hei
 
     if(list0){
         Picture *ref= &h->ref_list[0][ h->ref_cache[0][ scan8[n] ] ];
-        mc_dir_part(h, ref, n, square, chroma_height, delta, 0,
+        mc_dir_part(h, ref, n, square, height, delta, 0,
                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                           qpix_op, chroma_op, pixel_shift, chroma444);
+                           qpix_op, chroma_op, pixel_shift, chroma_idc);
 
         qpix_op=  qpix_avg;
         chroma_op= chroma_avg;
@@ -556,28 +577,36 @@ static inline void mc_part_std(H264Context *h, int n, int square, int chroma_hei
 
     if(list1){
         Picture *ref= &h->ref_list[1][ h->ref_cache[1][ scan8[n] ] ];
-        mc_dir_part(h, ref, n, square, chroma_height, delta, 1,
+        mc_dir_part(h, ref, n, square, height, delta, 1,
                            dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                           qpix_op, chroma_op, pixel_shift, chroma444);
+                           qpix_op, chroma_op, pixel_shift, chroma_idc);
     }
 }
 
-static inline void mc_part_weighted(H264Context *h, int n, int square, int chroma_height, int delta,
-                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
-                           int x_offset, int y_offset,
-                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
-                           h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
-                           h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
-                           int list0, int list1, int pixel_shift, int chroma444){
+static av_always_inline void
+mc_part_weighted(H264Context *h, int n, int square, int height, int delta,
+                 uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+                 int x_offset, int y_offset,
+                 qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
+                 h264_weight_func luma_weight_op, h264_weight_func chroma_weight_op,
+                 h264_biweight_func luma_weight_avg, h264_biweight_func chroma_weight_avg,
+                 int list0, int list1, int pixel_shift, int chroma_idc){
     MpegEncContext * const s = &h->s;
+    int chroma_height;
 
     dest_y += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
-    if(chroma444){
+    if (chroma_idc == 3 /* yuv444 */) {
+        chroma_height = height;
         chroma_weight_avg = luma_weight_avg;
         chroma_weight_op = luma_weight_op;
         dest_cb += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
         dest_cr += (2*x_offset << pixel_shift) + 2*y_offset*h->mb_linesize;
-    }else{
+    } else if (chroma_idc == 2 /* yuv422 */) {
+        chroma_height = height;
+        dest_cb += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
+        dest_cr += (  x_offset << pixel_shift) + 2*y_offset*h->mb_uvlinesize;
+    } else /* yuv420 */ {
+        chroma_height = height >> 1;
         dest_cb += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
         dest_cr += (  x_offset << pixel_shift) +   y_offset*h->mb_uvlinesize;
     }
@@ -593,27 +622,32 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
         int refn0 = h->ref_cache[0][ scan8[n] ];
         int refn1 = h->ref_cache[1][ scan8[n] ];
 
-        mc_dir_part(h, &h->ref_list[0][refn0], n, square, chroma_height, delta, 0,
+        mc_dir_part(h, &h->ref_list[0][refn0], n, square, height, delta, 0,
                     dest_y, dest_cb, dest_cr,
-                    x_offset, y_offset, qpix_put, chroma_put, pixel_shift, chroma444);
-        mc_dir_part(h, &h->ref_list[1][refn1], n, square, chroma_height, delta, 1,
+                    x_offset, y_offset, qpix_put, chroma_put,
+                    pixel_shift, chroma_idc);
+        mc_dir_part(h, &h->ref_list[1][refn1], n, square, height, delta, 1,
                     tmp_y, tmp_cb, tmp_cr,
-                    x_offset, y_offset, qpix_put, chroma_put, pixel_shift, chroma444);
+                    x_offset, y_offset, qpix_put, chroma_put,
+                    pixel_shift, chroma_idc);
 
         if(h->use_weight == 2){
             int weight0 = h->implicit_weight[refn0][refn1][s->mb_y&1];
             int weight1 = 64 - weight0;
-            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize, 5, weight0, weight1, 0);
-            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, 5, weight0, weight1, 0);
-            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, 5, weight0, weight1, 0);
+            luma_weight_avg(  dest_y,  tmp_y,  h->  mb_linesize,
+                              height,        5, weight0, weight1, 0);
+            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize,
+                              chroma_height, 5, weight0, weight1, 0);
+            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize,
+                              chroma_height, 5, weight0, weight1, 0);
         }else{
-            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, h->luma_log2_weight_denom,
+            luma_weight_avg(dest_y, tmp_y, h->mb_linesize, height, h->luma_log2_weight_denom,
                             h->luma_weight[refn0][0][0] , h->luma_weight[refn1][1][0],
                             h->luma_weight[refn0][0][1] + h->luma_weight[refn1][1][1]);
-            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
+            chroma_weight_avg(dest_cb, tmp_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
                             h->chroma_weight[refn0][0][0][0] , h->chroma_weight[refn1][1][0][0],
                             h->chroma_weight[refn0][0][0][1] + h->chroma_weight[refn1][1][0][1]);
-            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
+            chroma_weight_avg(dest_cr, tmp_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
                             h->chroma_weight[refn0][0][1][0] , h->chroma_weight[refn1][1][1][0],
                             h->chroma_weight[refn0][0][1][1] + h->chroma_weight[refn1][1][1][1]);
         }
@@ -621,42 +655,46 @@ static inline void mc_part_weighted(H264Context *h, int n, int square, int chrom
         int list = list1 ? 1 : 0;
         int refn = h->ref_cache[list][ scan8[n] ];
         Picture *ref= &h->ref_list[list][refn];
-        mc_dir_part(h, ref, n, square, chroma_height, delta, list,
+        mc_dir_part(h, ref, n, square, height, delta, list,
                     dest_y, dest_cb, dest_cr, x_offset, y_offset,
-                    qpix_put, chroma_put, pixel_shift, chroma444);
+                    qpix_put, chroma_put, pixel_shift, chroma_idc);
 
-        luma_weight_op(dest_y, h->mb_linesize, h->luma_log2_weight_denom,
+        luma_weight_op(dest_y, h->mb_linesize, height, h->luma_log2_weight_denom,
                        h->luma_weight[refn][list][0], h->luma_weight[refn][list][1]);
         if(h->use_weight_chroma){
-            chroma_weight_op(dest_cb, h->mb_uvlinesize, h->chroma_log2_weight_denom,
+            chroma_weight_op(dest_cb, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
                              h->chroma_weight[refn][list][0][0], h->chroma_weight[refn][list][0][1]);
-            chroma_weight_op(dest_cr, h->mb_uvlinesize, h->chroma_log2_weight_denom,
+            chroma_weight_op(dest_cr, h->mb_uvlinesize, chroma_height, h->chroma_log2_weight_denom,
                              h->chroma_weight[refn][list][1][0], h->chroma_weight[refn][list][1][1]);
         }
     }
 }
 
-static inline void mc_part(H264Context *h, int n, int square, int chroma_height, int delta,
-                           uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
-                           int x_offset, int y_offset,
-                           qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
-                           qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
-                           h264_weight_func *weight_op, h264_biweight_func *weight_avg,
-                           int list0, int list1, int pixel_shift, int chroma444){
+static av_always_inline void
+mc_part(H264Context *h, int n, int square, int height, int delta,
+        uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+        int x_offset, int y_offset,
+        qpel_mc_func *qpix_put, h264_chroma_mc_func chroma_put,
+        qpel_mc_func *qpix_avg, h264_chroma_mc_func chroma_avg,
+        h264_weight_func *weight_op, h264_biweight_func *weight_avg,
+        int list0, int list1, int pixel_shift, int chroma_idc)
+{
     if((h->use_weight==2 && list0 && list1
         && (h->implicit_weight[ h->ref_cache[0][scan8[n]] ][ h->ref_cache[1][scan8[n]] ][h->s.mb_y&1] != 32))
        || h->use_weight==1)
-        mc_part_weighted(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
+        mc_part_weighted(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
                          x_offset, y_offset, qpix_put, chroma_put,
-                         weight_op[0], weight_op[3], weight_avg[0],
-                         weight_avg[3], list0, list1, pixel_shift, chroma444);
+                         weight_op[0], weight_op[1], weight_avg[0],
+                         weight_avg[1], list0, list1, pixel_shift, chroma_idc);
     else
-        mc_part_std(h, n, square, chroma_height, delta, dest_y, dest_cb, dest_cr,
+        mc_part_std(h, n, square, height, delta, dest_y, dest_cb, dest_cr,
                     x_offset, y_offset, qpix_put, chroma_put, qpix_avg,
-                    chroma_avg, list0, list1, pixel_shift, chroma444);
+                    chroma_avg, list0, list1, pixel_shift, chroma_idc);
 }
 
-static inline void prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma444){
+static av_always_inline void
+prefetch_motion(H264Context *h, int list, int pixel_shift, int chroma_idc)
+{
     /* fetch pixels for estimated mv 4 macroblocks ahead
      * optimized for 64byte cache lines */
     MpegEncContext * const s = &h->s;
@@ -667,7 +705,7 @@ static inline void prefetch_motion(H264Context *h, int list, int pixel_shift, in
         uint8_t **src = h->ref_list[list][refn].f.data;
         int off= (mx << pixel_shift) + (my + (s->mb_x&3)*4)*h->mb_linesize + (64 << pixel_shift);
         s->dsp.prefetch(src[0]+off, s->linesize, 4);
-        if(chroma444){
+        if (chroma_idc == 3 /* yuv444 */) {
             s->dsp.prefetch(src[1]+off, s->linesize, 4);
             s->dsp.prefetch(src[2]+off, s->linesize, 4);
         }else{
@@ -681,45 +719,46 @@ static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t
                       qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
                       qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
                       h264_weight_func *weight_op, h264_biweight_func *weight_avg,
-                      int pixel_shift, int chroma444){
+                      int pixel_shift, int chroma_idc)
+{
     MpegEncContext * const s = &h->s;
     const int mb_xy= h->mb_xy;
     const int mb_type = s->current_picture.f.mb_type[mb_xy];
 
     assert(IS_INTER(mb_type));
 
-    if(HAVE_PTHREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
+    if(HAVE_THREADS && (s->avctx->active_thread_type & FF_THREAD_FRAME))
         await_references(h);
-    prefetch_motion(h, 0, pixel_shift, chroma444);
+    prefetch_motion(h, 0, pixel_shift, chroma_idc);
 
     if(IS_16X16(mb_type)){
-        mc_part(h, 0, 1, 8, 0, dest_y, dest_cb, dest_cr, 0, 0,
+        mc_part(h, 0, 1, 16, 0, dest_y, dest_cb, dest_cr, 0, 0,
                 qpix_put[0], chroma_put[0], qpix_avg[0], chroma_avg[0],
                 weight_op, weight_avg,
                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
-                pixel_shift, chroma444);
+                pixel_shift, chroma_idc);
     }else if(IS_16X8(mb_type)){
-        mc_part(h, 0, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
+        mc_part(h, 0, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 0,
                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
-                &weight_op[1], &weight_avg[1],
+                weight_op, weight_avg,
                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
-                pixel_shift, chroma444);
-        mc_part(h, 8, 0, 4, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
+                pixel_shift, chroma_idc);
+        mc_part(h, 8, 0, 8, 8 << pixel_shift, dest_y, dest_cb, dest_cr, 0, 4,
                 qpix_put[1], chroma_put[0], qpix_avg[1], chroma_avg[0],
-                &weight_op[1], &weight_avg[1],
+                weight_op, weight_avg,
                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
-                pixel_shift, chroma444);
+                pixel_shift, chroma_idc);
     }else if(IS_8X16(mb_type)){
-        mc_part(h, 0, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
+        mc_part(h, 0, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 0, 0,
                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
-                &weight_op[2], &weight_avg[2],
+                &weight_op[1], &weight_avg[1],
                 IS_DIR(mb_type, 0, 0), IS_DIR(mb_type, 0, 1),
-                pixel_shift, chroma444);
-        mc_part(h, 4, 0, 8, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
+                pixel_shift, chroma_idc);
+        mc_part(h, 4, 0, 16, 8*h->mb_linesize, dest_y, dest_cb, dest_cr, 4, 0,
                 qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
-                &weight_op[2], &weight_avg[2],
+                &weight_op[1], &weight_avg[1],
                 IS_DIR(mb_type, 1, 0), IS_DIR(mb_type, 1, 1),
-                pixel_shift, chroma444);
+                pixel_shift, chroma_idc);
     }else{
         int i;
 
@@ -732,50 +771,72 @@ static av_always_inline void hl_motion(H264Context *h, uint8_t *dest_y, uint8_t
             int y_offset= (i&2)<<1;
 
             if(IS_SUB_8X8(sub_mb_type)){
-                mc_part(h, n, 1, 4, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
+                mc_part(h, n, 1, 8, 0, dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_put[1], chroma_put[1], qpix_avg[1], chroma_avg[1],
-                    &weight_op[3], &weight_avg[3],
+                    &weight_op[1], &weight_avg[1],
                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
-                    pixel_shift, chroma444);
+                    pixel_shift, chroma_idc);
             }else if(IS_SUB_8X4(sub_mb_type)){
-                mc_part(h, n  , 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
+                mc_part(h, n  , 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
-                    &weight_op[4], &weight_avg[4],
+                    &weight_op[1], &weight_avg[1],
                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
-                    pixel_shift, chroma444);
-                mc_part(h, n+2, 0, 2, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
+                    pixel_shift, chroma_idc);
+                mc_part(h, n+2, 0, 4, 4 << pixel_shift, dest_y, dest_cb, dest_cr, x_offset, y_offset+2,
                     qpix_put[2], chroma_put[1], qpix_avg[2], chroma_avg[1],
-                    &weight_op[4], &weight_avg[4],
+                    &weight_op[1], &weight_avg[1],
                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
-                    pixel_shift, chroma444);
+                    pixel_shift, chroma_idc);
             }else if(IS_SUB_4X8(sub_mb_type)){
-                mc_part(h, n  , 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
+                mc_part(h, n  , 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset, y_offset,
                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
-                    &weight_op[5], &weight_avg[5],
+                    &weight_op[2], &weight_avg[2],
                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
-                    pixel_shift, chroma444);
-                mc_part(h, n+1, 0, 4, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
+                    pixel_shift, chroma_idc);
+                mc_part(h, n+1, 0, 8, 4*h->mb_linesize, dest_y, dest_cb, dest_cr, x_offset+2, y_offset,
                     qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
-                    &weight_op[5], &weight_avg[5],
+                    &weight_op[2], &weight_avg[2],
                     IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
-                    pixel_shift, chroma444);
+                    pixel_shift, chroma_idc);
             }else{
                 int j;
                 assert(IS_SUB_4X4(sub_mb_type));
                 for(j=0; j<4; j++){
                     int sub_x_offset= x_offset + 2*(j&1);
                     int sub_y_offset= y_offset +   (j&2);
-                    mc_part(h, n+j, 1, 2, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
+                    mc_part(h, n+j, 1, 4, 0, dest_y, dest_cb, dest_cr, sub_x_offset, sub_y_offset,
                         qpix_put[2], chroma_put[2], qpix_avg[2], chroma_avg[2],
-                        &weight_op[6], &weight_avg[6],
+                        &weight_op[2], &weight_avg[2],
                         IS_DIR(sub_mb_type, 0, 0), IS_DIR(sub_mb_type, 0, 1),
-                        pixel_shift, chroma444);
+                        pixel_shift, chroma_idc);
                 }
             }
         }
     }
 
-    prefetch_motion(h, 1, pixel_shift, chroma444);
+    prefetch_motion(h, 1, pixel_shift, chroma_idc);
+}
+
+static av_always_inline void
+hl_motion_420(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+              qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
+              qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
+              h264_weight_func *weight_op, h264_biweight_func *weight_avg,
+              int pixel_shift)
+{
+    hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
+              qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 1);
+}
+
+static av_always_inline void
+hl_motion_422(H264Context *h, uint8_t *dest_y, uint8_t *dest_cb, uint8_t *dest_cr,
+              qpel_mc_func (*qpix_put)[16], h264_chroma_mc_func (*chroma_put),
+              qpel_mc_func (*qpix_avg)[16], h264_chroma_mc_func (*chroma_avg),
+              h264_weight_func *weight_op, h264_biweight_func *weight_avg,
+              int pixel_shift)
+{
+    hl_motion(h, dest_y, dest_cb, dest_cr, qpix_put, chroma_put,
+              qpix_avg, chroma_avg, weight_op, weight_avg, pixel_shift, 2);
 }
 
 static void free_tables(H264Context *h, int free_rbsp){
@@ -942,7 +1003,7 @@ static void clone_tables(H264Context *dst, H264Context *src, int i){
     dst->list_counts              = src->list_counts;
 
     dst->s.obmc_scratchpad = NULL;
-    ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma);
+    ff_h264_pred_init(&dst->hpc, src->s.codec_id, src->sps.bit_depth_luma, src->sps.chroma_format_idc);
 }
 
 /**
@@ -970,12 +1031,11 @@ static av_cold void common_init(H264Context *h){
     s->height = s->avctx->height;
     s->codec_id= s->avctx->codec->id;
 
-    ff_h264dsp_init(&h->h264dsp, 8);
-    ff_h264_pred_init(&h->hpc, s->codec_id, 8);
+    ff_h264dsp_init(&h->h264dsp, 8, 1);
+    ff_h264_pred_init(&h->hpc, s->codec_id, 8, 1);
 
     h->dequant_coeff_pps= -1;
     s->unrestricted_mv=1;
-    s->decode=1; //FIXME
 
     dsputil_init(&s->dsp, s->avctx); // needed so that idct permutation is known early
 
@@ -1005,6 +1065,8 @@ int ff_h264_decode_extradata(H264Context *h)
         p += 6;
         for (i = 0; i < cnt; i++) {
             nalsize = AV_RB16(p) + 2;
+            if (p - avctx->extradata + nalsize > avctx->extradata_size)
+                return -1;
             if(decode_nal_units(h, p, nalsize) < 0) {
                 av_log(avctx, AV_LOG_ERROR, "Decoding sps %d from avcC failed\n", i);
                 return -1;
@@ -1015,6 +1077,8 @@ int ff_h264_decode_extradata(H264Context *h)
         cnt = *(p++); // Number of pps
         for (i = 0; i < cnt; i++) {
             nalsize = AV_RB16(p) + 2;
+            if (p - avctx->extradata + nalsize > avctx->extradata_size)
+                return -1;
             if (decode_nal_units(h, p, nalsize) < 0) {
                 av_log(avctx, AV_LOG_ERROR, "Decoding pps %d from avcC failed\n", i);
                 return -1;
@@ -1034,6 +1098,7 @@ int ff_h264_decode_extradata(H264Context *h)
 av_cold int ff_h264_decode_init(AVCodecContext *avctx){
     H264Context *h= avctx->priv_data;
     MpegEncContext * const s = &h->s;
+    int i;
 
     MPV_decode_defaults(s);
 
@@ -1058,6 +1123,8 @@ av_cold int ff_h264_decode_init(AVCodecContext *avctx){
 
     h->thread_context[0] = h;
     h->outputed_poc = h->next_outputed_poc = INT_MIN;
+    for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
+        h->last_pocs[i] = INT_MIN;
     h->prev_poc_msb= 1<<16;
     h->x264_build = -1;
     ff_h264_reset_sei(h);
@@ -1108,7 +1175,8 @@ static void copy_parameter_set(void **to, void **from, int count, int size)
 static int decode_init_thread_copy(AVCodecContext *avctx){
     H264Context *h= avctx->priv_data;
 
-    if (!avctx->is_copy) return 0;
+    if (!avctx->internal->is_copy)
+        return 0;
     memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
     memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
 
@@ -1138,7 +1206,10 @@ static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContex
         memcpy(&h->s + 1, &h1->s + 1, sizeof(H264Context) - sizeof(MpegEncContext)); //copy all fields after MpegEnc
         memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
         memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
-        ff_h264_alloc_tables(h);
+        if (ff_h264_alloc_tables(h) < 0) {
+            av_log(dst, AV_LOG_ERROR, "Could not allocate memory for h264\n");
+            return AVERROR(ENOMEM);
+        }
         context_init(h);
 
         for(i=0; i<2; i++){
@@ -1194,7 +1265,7 @@ static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContex
     if(!s->current_picture_ptr) return 0;
 
     if(!s->dropable) {
-        ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
+        err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
         h->prev_poc_msb     = h->poc_msb;
         h->prev_poc_lsb     = h->poc_lsb;
     }
@@ -1202,14 +1273,13 @@ static int decode_update_thread_context(AVCodecContext *dst, const AVCodecContex
     h->prev_frame_num       = h->frame_num;
     h->outputed_poc         = h->next_outputed_poc;
 
-    return 0;
+    return err;
 }
 
 int ff_h264_frame_start(H264Context *h){
     MpegEncContext * const s = &h->s;
     int i;
     const int pixel_shift = h->pixel_shift;
-    int thread_count = (s->avctx->active_thread_type & FF_THREAD_SLICE) ? s->avctx->thread_count : 1;
 
     if(MPV_frame_start(s, s->avctx) < 0)
         return -1;
@@ -1238,7 +1308,7 @@ int ff_h264_frame_start(H264Context *h){
 
     /* can't be in alloc_tables because linesize isn't known there.
      * FIXME: redo bipred weight to not require extra buffer? */
-    for(i = 0; i < thread_count; i++)
+    for(i = 0; i < s->slice_context_count; i++)
         if(h->thread_context[i] && !h->thread_context[i]->s.obmc_scratchpad)
             h->thread_context[i]->s.obmc_scratchpad = av_malloc(16*6*s->linesize);
 
@@ -1279,6 +1349,7 @@ static void decode_postinit(H264Context *h, int setup_finished){
     Picture *out = s->current_picture_ptr;
     Picture *cur = s->current_picture_ptr;
     int i, pics, out_of_order, out_idx;
+    int invalid = 0, cnt = 0;
 
     s->current_picture_ptr->f.qscale_type = FF_QSCALE_TYPE_H264;
     s->current_picture_ptr->f.pict_type   = s->pict_type;
@@ -1369,7 +1440,7 @@ static void decode_postinit(H264Context *h, int setup_finished){
 
     if(   s->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT
        && !h->sps.bitstream_restriction_flag){
-        s->avctx->has_b_frames= MAX_DELAYED_PIC_COUNT;
+        s->avctx->has_b_frames = MAX_DELAYED_PIC_COUNT - 1;
         s->low_delay= 0;
     }
 
@@ -1382,41 +1453,89 @@ static void decode_postinit(H264Context *h, int setup_finished){
     if (cur->f.reference == 0)
         cur->f.reference = DELAYED_PIC_REF;
 
+    /* Frame reordering. This code takes pictures from coding order and sorts
+     * them by their incremental POC value into display order. It supports POC
+     * gaps, MMCO reset codes and random resets.
+     * A "display group" can start either with a IDR frame (f.key_frame = 1),
+     * and/or can be closed down with a MMCO reset code. In sequences where
+     * there is no delay, we can't detect that (since the frame was already
+     * output to the user), so we also set h->mmco_reset to detect the MMCO
+     * reset code.
+     * FIXME: if we detect insufficient delays (as per s->avctx->has_b_frames),
+     * we increase the delay between input and output. All frames affected by
+     * the lag (e.g. those that should have been output before another frame
+     * that we already returned to the user) will be dropped. This is a bug
+     * that we will fix later. */
+    for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++) {
+        cnt     += out->poc < h->last_pocs[i];
+        invalid += out->poc == INT_MIN;
+    }
+    if (!h->mmco_reset && !cur->f.key_frame && cnt + invalid == MAX_DELAYED_PIC_COUNT && cnt > 0) {
+        h->mmco_reset = 2;
+        if (pics > 1)
+            h->delayed_pic[pics - 2]->mmco_reset = 2;
+    }
+    if (h->mmco_reset || cur->f.key_frame) {
+        for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
+            h->last_pocs[i] = INT_MIN;
+        cnt     = 0;
+        invalid = MAX_DELAYED_PIC_COUNT;
+    }
     out = h->delayed_pic[0];
     out_idx = 0;
-    for (i = 1; h->delayed_pic[i] && !h->delayed_pic[i]->f.key_frame && !h->delayed_pic[i]->mmco_reset; i++)
+    for (i = 1; i < MAX_DELAYED_PIC_COUNT && h->delayed_pic[i] &&
+         !h->delayed_pic[i-1]->mmco_reset && !h->delayed_pic[i]->f.key_frame; i++)
+    {
         if(h->delayed_pic[i]->poc < out->poc){
             out = h->delayed_pic[i];
             out_idx = i;
         }
-    if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset))
-        h->next_outputed_poc= INT_MIN;
-    out_of_order = out->poc < h->next_outputed_poc;
+    }
+    if (s->avctx->has_b_frames == 0 && (h->delayed_pic[0]->f.key_frame || h->mmco_reset))
+        h->next_outputed_poc = INT_MIN;
+    out_of_order = !out->f.key_frame && !h->mmco_reset && (out->poc < h->next_outputed_poc);
 
     if(h->sps.bitstream_restriction_flag && s->avctx->has_b_frames >= h->sps.num_reorder_frames)
         { }
-    else if((out_of_order && pics-1 == s->avctx->has_b_frames && s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT)
-       || (s->low_delay &&
-        ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2)
-         || cur->f.pict_type == AV_PICTURE_TYPE_B)))
-    {
+    else if (out_of_order && pics-1 == s->avctx->has_b_frames &&
+             s->avctx->has_b_frames < MAX_DELAYED_PIC_COUNT) {
+        if (invalid + cnt < MAX_DELAYED_PIC_COUNT) {
+            s->avctx->has_b_frames = FFMAX(s->avctx->has_b_frames, cnt);
+        }
+        s->low_delay = 0;
+    } else if (s->low_delay &&
+               ((h->next_outputed_poc != INT_MIN && out->poc > h->next_outputed_poc + 2) ||
+                cur->f.pict_type == AV_PICTURE_TYPE_B)) {
         s->low_delay = 0;
         s->avctx->has_b_frames++;
     }
 
-    if(out_of_order || pics > s->avctx->has_b_frames){
+    if(pics > s->avctx->has_b_frames){
         out->f.reference &= ~DELAYED_PIC_REF;
         out->owner2 = s; // for frame threading, the owner must be the second field's thread
                          // or else the first thread can release the picture and reuse it unsafely
         for(i=out_idx; h->delayed_pic[i]; i++)
             h->delayed_pic[i] = h->delayed_pic[i+1];
     }
+    memmove(h->last_pocs, &h->last_pocs[1], sizeof(*h->last_pocs) * (MAX_DELAYED_PIC_COUNT - 1));
+    h->last_pocs[MAX_DELAYED_PIC_COUNT - 1] = cur->poc;
     if(!out_of_order && pics > s->avctx->has_b_frames){
         h->next_output_pic = out;
-        if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f.key_frame || h->delayed_pic[0]->mmco_reset)) {
-            h->next_outputed_poc = INT_MIN;
-        } else
-            h->next_outputed_poc = out->poc;
+        if (out->mmco_reset) {
+            if (out_idx > 0) {
+                h->next_outputed_poc = out->poc;
+                h->delayed_pic[out_idx - 1]->mmco_reset = out->mmco_reset;
+            } else {
+                h->next_outputed_poc = INT_MIN;
+            }
+        } else {
+            if (out_idx == 0 && pics > 1 && h->delayed_pic[0]->f.key_frame) {
+                h->next_outputed_poc = INT_MIN;
+            } else {
+                h->next_outputed_poc = out->poc;
+            }
+        }
+        h->mmco_reset = 0;
     }else{
         av_log(s->avctx, AV_LOG_DEBUG, "no picture\n");
     }
@@ -1425,11 +1544,16 @@ static void decode_postinit(H264Context *h, int setup_finished){
         ff_thread_finish_setup(s->avctx);
 }
 
-static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y, uint8_t *src_cb, uint8_t *src_cr, int linesize, int uvlinesize, int chroma444, int simple){
+static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
+                                              uint8_t *src_cb, uint8_t *src_cr,
+                                              int linesize, int uvlinesize, int simple)
+{
     MpegEncContext * const s = &h->s;
     uint8_t *top_border;
     int top_idx = 1;
     const int pixel_shift = h->pixel_shift;
+    int chroma444 = CHROMA444;
+    int chroma422 = CHROMA422;
 
     src_y  -=   linesize;
     src_cb -= uvlinesize;
@@ -1453,6 +1577,14 @@ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y, ui
                             AV_COPY128(top_border+16, src_cb + 15*uvlinesize);
                             AV_COPY128(top_border+32, src_cr + 15*uvlinesize);
                         }
+                    } else if(chroma422) {
+                        if (pixel_shift) {
+                            AV_COPY128(top_border+32, src_cb + 15*uvlinesize);
+                            AV_COPY128(top_border+48, src_cr + 15*uvlinesize);
+                        } else {
+                            AV_COPY64(top_border+16, src_cb +  15*uvlinesize);
+                            AV_COPY64(top_border+24, src_cr +  15*uvlinesize);
+                        }
                     } else {
                         if (pixel_shift) {
                             AV_COPY128(top_border+32, src_cb+7*uvlinesize);
@@ -1488,6 +1620,14 @@ static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y, ui
                 AV_COPY128(top_border+16, src_cb + 16*linesize);
                 AV_COPY128(top_border+32, src_cr + 16*linesize);
             }
+        } else if(chroma422) {
+            if (pixel_shift) {
+                AV_COPY128(top_border+32, src_cb+16*uvlinesize);
+                AV_COPY128(top_border+48, src_cr+16*uvlinesize);
+            } else {
+                AV_COPY64(top_border+16, src_cb+16*uvlinesize);
+                AV_COPY64(top_border+24, src_cr+16*uvlinesize);
+            }
         } else {
             if (pixel_shift) {
                 AV_COPY128(top_border+32, src_cb+8*uvlinesize);
@@ -1661,7 +1801,7 @@ static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_ty
                                     tr_high= ((uint16_t*)ptr)[3 - linesize/2]*0x0001000100010001ULL;
                                     topright= (uint8_t*) &tr_high;
                                 } else {
-                                    tr= ptr[3 - linesize]*0x01010101;
+                                    tr= ptr[3 - linesize]*0x01010101u;
                                     topright= (uint8_t*) &tr;
                                 }
                             }else
@@ -1677,7 +1817,7 @@ static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_ty
                                     idct_dc_add(ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
                                 else
                                     idct_add   (ptr, h->mb + (i*16+p*256 << pixel_shift), linesize);
-                            }else
+                            } else if (CONFIG_SVQ3_DECODER)
                                 ff_svq3_add_idct_c(ptr, h->mb + i*16+p*256, linesize, qscale, 0);
                         }
                     }
@@ -1694,10 +1834,10 @@ static av_always_inline void hl_decode_mb_predict_luma(H264Context *h, int mb_ty
                     static const uint8_t dc_mapping[16] = { 0*16, 1*16, 4*16, 5*16, 2*16, 3*16, 6*16, 7*16,
                                                             8*16, 9*16,12*16,13*16,10*16,11*16,14*16,15*16};
                     for(i = 0; i < 16; i++)
-                        dctcoef_set(h->mb+p*256, pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
+                        dctcoef_set(h->mb+(p*256 << pixel_shift), pixel_shift, dc_mapping[i], dctcoef_get(h->mb_luma_dc[p], pixel_shift, i));
                 }
             }
-        }else
+        } else if (CONFIG_SVQ3_DECODER)
             ff_svq3_luma_dc_dequant_idct_c(h->mb+p*256, h->mb_luma_dc[p], qscale);
     }
 }
@@ -1741,7 +1881,7 @@ static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
                     }
                 }
             }
-        }else{
+        } else if (CONFIG_SVQ3_DECODER) {
             for(i=0; i<16; i++){
                 if(h->non_zero_count_cache[ scan8[i+p*16] ] || h->mb[i*16+p*256]){ //FIXME benchmark weird rule, & below
                     uint8_t * const ptr= dest_y + block_offset[i];
@@ -1752,7 +1892,8 @@ static av_always_inline void hl_decode_mb_idct_luma(H264Context *h, int mb_type,
     }
 }
 
-static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift){
+static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, int pixel_shift)
+{
     MpegEncContext * const s = &h->s;
     const int mb_x= s->mb_x;
     const int mb_y= s->mb_y;
@@ -1766,10 +1907,12 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
     /* is_h264 should always be true if SVQ3 is disabled. */
     const int is_h264 = !CONFIG_SVQ3_DECODER || simple || s->codec_id == CODEC_ID_H264;
     void (*idct_add)(uint8_t *dst, DCTELEM *block, int stride);
+    const int block_h = 16 >> s->chroma_y_shift;
+    const int chroma422 = CHROMA422;
 
     dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
-    dest_cb = s->current_picture.f.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) *  8;
-    dest_cr = s->current_picture.f.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) *  8;
+    dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
+    dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift)*8 + mb_y * s->uvlinesize * block_h;
 
     s->dsp.prefetch(dest_y + (s->mb_x&3)*4*s->linesize + (64 << pixel_shift), s->linesize, 4);
     s->dsp.prefetch(dest_cb + (s->mb_x&7)*s->uvlinesize + (64 << pixel_shift), dest_cr - dest_cb, 2);
@@ -1782,8 +1925,8 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
         block_offset = &h->block_offset[48];
         if(mb_y&1){ //FIXME move out of this function?
             dest_y -= s->linesize*15;
-            dest_cb-= s->uvlinesize*7;
-            dest_cr-= s->uvlinesize*7;
+            dest_cb-= s->uvlinesize * (block_h - 1);
+            dest_cr-= s->uvlinesize * (block_h - 1);
         }
         if(FRAME_MBAFF) {
             int list;
@@ -1821,15 +1964,30 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
                     tmp_y[j] = get_bits(&gb, bit_depth);
             }
             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-                for (i = 0; i < 8; i++) {
-                    uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
-                    for (j = 0; j < 8; j++)
-                        tmp_cb[j] = get_bits(&gb, bit_depth);
-                }
-                for (i = 0; i < 8; i++) {
-                    uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
-                    for (j = 0; j < 8; j++)
-                        tmp_cr[j] = get_bits(&gb, bit_depth);
+                if (!h->sps.chroma_format_idc) {
+                    for (i = 0; i < block_h; i++) {
+                        uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
+                        for (j = 0; j < 8; j++) {
+                            tmp_cb[j] = 1 << (bit_depth - 1);
+                        }
+                    }
+                    for (i = 0; i < block_h; i++) {
+                        uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
+                        for (j = 0; j < 8; j++) {
+                            tmp_cr[j] = 1 << (bit_depth - 1);
+                        }
+                    }
+                } else {
+                    for (i = 0; i < block_h; i++) {
+                        uint16_t *tmp_cb = (uint16_t*)(dest_cb + i*uvlinesize);
+                        for (j = 0; j < 8; j++)
+                            tmp_cb[j] = get_bits(&gb, bit_depth);
+                    }
+                    for (i = 0; i < block_h; i++) {
+                        uint16_t *tmp_cr = (uint16_t*)(dest_cr + i*uvlinesize);
+                        for (j = 0; j < 8; j++)
+                            tmp_cr[j] = get_bits(&gb, bit_depth);
+                    }
                 }
             }
         } else {
@@ -1837,9 +1995,16 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
                 memcpy(dest_y + i*  linesize, h->mb       + i*8, 16);
             }
             if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
-                for (i=0; i<8; i++) {
-                    memcpy(dest_cb+ i*uvlinesize, h->mb + 128 + i*4,  8);
-                    memcpy(dest_cr+ i*uvlinesize, h->mb + 160 + i*4,  8);
+                if (!h->sps.chroma_format_idc) {
+                    for (i = 0; i < block_h; i++) {
+                        memset(dest_cb + i*uvlinesize, 128, 8);
+                        memset(dest_cr + i*uvlinesize, 128, 8);
+                    }
+                } else {
+                    for (i = 0; i < block_h; i++) {
+                        memcpy(dest_cb + i*uvlinesize, h->mb + 128 + i*4,  8);
+                        memcpy(dest_cr + i*uvlinesize, h->mb + 160 + i*4,  8);
+                    }
                 }
             }
         }
@@ -1858,11 +2023,21 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
             if(h->deblocking_filter)
                 xchg_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0, 0, simple, pixel_shift);
         }else if(is_h264){
-            hl_motion(h, dest_y, dest_cb, dest_cr,
-                      s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
-                      s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
-                      h->h264dsp.weight_h264_pixels_tab,
-                      h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 0);
+            if (chroma422) {
+                hl_motion_422(h, dest_y, dest_cb, dest_cr,
+                              s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
+                              s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
+                              h->h264dsp.weight_h264_pixels_tab,
+                              h->h264dsp.biweight_h264_pixels_tab,
+                              pixel_shift);
+            } else {
+                hl_motion_420(h, dest_y, dest_cb, dest_cr,
+                              s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
+                              s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
+                              h->h264dsp.weight_h264_pixels_tab,
+                              h->h264dsp.biweight_h264_pixels_tab,
+                              pixel_shift);
+            }
         }
 
         hl_decode_mb_idct_luma(h, mb_type, is_h264, simple, transform_bypass, pixel_shift, block_offset, linesize, dest_y, 0);
@@ -1880,18 +2055,32 @@ static av_always_inline void hl_decode_mb_internal(H264Context *h, int simple, i
                             if(h->non_zero_count_cache[ scan8[i] ] || dctcoef_get(h->mb, pixel_shift, i*16))
                                 idct_add   (dest[j-1] + block_offset[i], h->mb + (i*16 << pixel_shift), uvlinesize);
                         }
+                        if (chroma422) {
+                            for(i=j*16+4; i<j*16+8; i++){
+                                if(h->non_zero_count_cache[ scan8[i+4] ] || dctcoef_get(h->mb, pixel_shift, i*16))
+                                    idct_add   (dest[j-1] + block_offset[i+4], h->mb + (i*16 << pixel_shift), uvlinesize);
+                            }
+                        }
                     }
                 }
             }else{
                 if(is_h264){
+                    int qp[2];
+                    if (chroma422) {
+                        qp[0] = h->chroma_qp[0] + 3;
+                        qp[1] = h->chroma_qp[1] + 3;
+                    } else {
+                        qp[0] = h->chroma_qp[0];
+                        qp[1] = h->chroma_qp[1];
+                    }
                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+0] ])
-                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
+                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*1 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][qp[0]][0]);
                     if(h->non_zero_count_cache[ scan8[CHROMA_DC_BLOCK_INDEX+1] ])
-                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
+                        h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + (16*16*2 << pixel_shift), h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][qp[1]][0]);
                     h->h264dsp.h264_idct_add8(dest, block_offset,
                                               h->mb, uvlinesize,
                                               h->non_zero_count_cache);
-                }else{
+                } else if (CONFIG_SVQ3_DECODER) {
                     h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*1, h->dequant4_coeff[IS_INTRA(mb_type) ? 1:4][h->chroma_qp[0]][0]);
                     h->h264dsp.h264_chroma_dc_dequant_idct(h->mb + 16*16*2, h->dequant4_coeff[IS_INTRA(mb_type) ? 2:5][h->chroma_qp[1]][0]);
                     for(j=1; j<3; j++){
@@ -1996,7 +2185,7 @@ static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simpl
                       s->me.qpel_put, s->dsp.put_h264_chroma_pixels_tab,
                       s->me.qpel_avg, s->dsp.avg_h264_chroma_pixels_tab,
                       h->h264dsp.weight_h264_pixels_tab,
-                      h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 1);
+                      h->h264dsp.biweight_h264_pixels_tab, pixel_shift, 3);
         }
 
         for (p = 0; p < plane_count; p++)
@@ -2016,8 +2205,8 @@ static av_always_inline void hl_decode_mb_444_internal(H264Context *h, int simpl
 static void hl_decode_mb_simple_ ## bits(H264Context *h){ \
     hl_decode_mb_internal(h, 1, sh); \
 }
-hl_decode_mb_simple(0, 8);
-hl_decode_mb_simple(1, 16);
+hl_decode_mb_simple(0, 8)
+hl_decode_mb_simple(1, 16)
 
 /**
  * Process a macroblock; this handles edge cases, such as interlacing.
@@ -2129,7 +2318,11 @@ static void implicit_weight_table(H264Context *h, int field){
     }
 
     if(field < 0){
-        cur_poc = s->current_picture_ptr->poc;
+        if (s->picture_structure == PICT_FRAME) {
+            cur_poc = s->current_picture_ptr->poc;
+        } else {
+            cur_poc = s->current_picture_ptr->field_poc[s->picture_structure - 1];
+        }
     if(   h->ref_count[0] == 1 && h->ref_count[1] == 1 && !FRAME_MBAFF
        && h->ref_list[0][0].poc + h->ref_list[1][0].poc == 2*cur_poc){
         h->use_weight= 0;
@@ -2154,15 +2347,17 @@ static void implicit_weight_table(H264Context *h, int field){
     for(ref0=ref_start; ref0 < ref_count0; ref0++){
         int poc0 = h->ref_list[0][ref0].poc;
         for(ref1=ref_start; ref1 < ref_count1; ref1++){
-            int poc1 = h->ref_list[1][ref1].poc;
-            int td = av_clip(poc1 - poc0, -128, 127);
-            int w= 32;
-            if(td){
-                int tb = av_clip(cur_poc - poc0, -128, 127);
-                int tx = (16384 + (FFABS(td) >> 1)) / td;
-                int dist_scale_factor = (tb*tx + 32) >> 8;
-                if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
-                    w = 64 - dist_scale_factor;
+            int w = 32;
+            if (!h->ref_list[0][ref0].long_ref && !h->ref_list[1][ref1].long_ref) {
+                int poc1 = h->ref_list[1][ref1].poc;
+                int td = av_clip(poc1 - poc0, -128, 127);
+                if(td){
+                    int tb = av_clip(cur_poc - poc0, -128, 127);
+                    int tx = (16384 + (FFABS(td) >> 1)) / td;
+                    int dist_scale_factor = (tb*tx + 32) >> 8;
+                    if(dist_scale_factor >= -64 && dist_scale_factor <= 128)
+                        w = 64 - dist_scale_factor;
+                }
             }
             if(field<0){
                 h->implicit_weight[ref0][ref1][0]=
@@ -2194,6 +2389,8 @@ static void flush_dpb(AVCodecContext *avctx){
             h->delayed_pic[i]->f.reference = 0;
         h->delayed_pic[i]= NULL;
     }
+    for (i = 0; i < MAX_DELAYED_PIC_COUNT; i++)
+        h->last_pocs[i] = INT_MIN;
     h->outputed_poc=h->next_outputed_poc= INT_MIN;
     h->prev_interlaced_frame = 1;
     idr(h);
@@ -2318,9 +2515,10 @@ static void init_scan_tables(H264Context *h){
     }
 }
 
-static void field_end(H264Context *h, int in_setup){
+static int field_end(H264Context *h, int in_setup){
     MpegEncContext * const s = &h->s;
     AVCodecContext * const avctx= s->avctx;
+    int err = 0;
     s->mb_y= 0;
 
     if (!in_setup && !s->dropable)
@@ -2332,7 +2530,7 @@ static void field_end(H264Context *h, int in_setup){
 
     if(in_setup || !(avctx->active_thread_type&FF_THREAD_FRAME)){
         if(!s->dropable) {
-            ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
+            err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
             h->prev_poc_msb= h->poc_msb;
             h->prev_poc_lsb= h->poc_lsb;
         }
@@ -2367,6 +2565,8 @@ static void field_end(H264Context *h, int in_setup){
     MPV_frame_end(s);
 
     h->current_slice=0;
+
+    return err;
 }
 
 /**
@@ -2397,7 +2597,7 @@ static void clone_slice(H264Context *dst, H264Context *src)
 }
 
 /**
- * computes profile from profile_idc and constraint_set?_flags
+ * Compute profile from profile_idc and constraint_set?_flags.
  *
  * @param sps SPS
  *
@@ -2424,7 +2624,7 @@ int ff_h264_get_profile(SPS *sps)
 }
 
 /**
- * decodes a slice header.
+ * Decode a slice header.
  * This will also call MPV_common_init() and frame_start() as needed.
  *
  * @param h h264context
@@ -2507,21 +2707,18 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     s->avctx->level   = h->sps.level_idc;
     s->avctx->refs    = h->sps.ref_frame_count;
 
-    if(h == h0 && h->dequant_coeff_pps != pps_id){
-        h->dequant_coeff_pps = pps_id;
-        init_dequant_tables(h);
-    }
-
     s->mb_width= h->sps.mb_width;
     s->mb_height= h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
 
     h->b_stride=  s->mb_width*4;
 
+    s->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
+
     s->width = 16*s->mb_width - (2>>CHROMA444)*FFMIN(h->sps.crop_right, (8<<CHROMA444)-1);
     if(h->sps.frame_mbs_only_flag)
-        s->height= 16*s->mb_height - (2>>CHROMA444)*FFMIN(h->sps.crop_bottom, (8<<CHROMA444)-1);
+        s->height= 16*s->mb_height - (1<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
     else
-        s->height= 16*s->mb_height - (4>>CHROMA444)*FFMIN(h->sps.crop_bottom, (8<<CHROMA444)-1);
+        s->height= 16*s->mb_height - (2<<s->chroma_y_shift)*FFMIN(h->sps.crop_bottom, (16>>s->chroma_y_shift)-1);
 
     if (s->context_initialized
         && (   s->width != s->avctx->width || s->height != s->avctx->height
@@ -2544,9 +2741,6 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         s->avctx->sample_aspect_ratio= h->sps.sar;
         av_assert0(s->avctx->sample_aspect_ratio.den);
 
-        h->s.avctx->coded_width = 16*s->mb_width;
-        h->s.avctx->coded_height = 16*s->mb_height;
-
         if(h->sps.video_signal_type_present_flag){
             s->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG : AVCOL_RANGE_MPEG;
             if(h->sps.colour_description_present_flag){
@@ -2566,14 +2760,35 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
 
         switch (h->sps.bit_depth_luma) {
             case 9 :
-                s->avctx->pix_fmt = CHROMA444 ? PIX_FMT_YUV444P9 : PIX_FMT_YUV420P9;
+                if (CHROMA444) {
+                    if (s->avctx->colorspace == AVCOL_SPC_RGB) {
+                        s->avctx->pix_fmt = PIX_FMT_GBRP9;
+                    } else
+                        s->avctx->pix_fmt = PIX_FMT_YUV444P9;
+                } else if (CHROMA422)
+                    s->avctx->pix_fmt = PIX_FMT_YUV422P9;
+                else
+                    s->avctx->pix_fmt = PIX_FMT_YUV420P9;
                 break;
             case 10 :
-                s->avctx->pix_fmt = CHROMA444 ? PIX_FMT_YUV444P10 : PIX_FMT_YUV420P10;
+                if (CHROMA444) {
+                    if (s->avctx->colorspace == AVCOL_SPC_RGB) {
+                        s->avctx->pix_fmt = PIX_FMT_GBRP10;
+                    } else
+                        s->avctx->pix_fmt = PIX_FMT_YUV444P10;
+                } else if (CHROMA422)
+                    s->avctx->pix_fmt = PIX_FMT_YUV422P10;
+                else
+                    s->avctx->pix_fmt = PIX_FMT_YUV420P10;
                 break;
-            default:
+            case 8:
                 if (CHROMA444){
-                    s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
+                    if (s->avctx->colorspace == AVCOL_SPC_RGB) {
+                        s->avctx->pix_fmt = PIX_FMT_GBRP;
+                    } else
+                        s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ444P : PIX_FMT_YUV444P;
+                } else if (CHROMA422) {
+                    s->avctx->pix_fmt = s->avctx->color_range == AVCOL_RANGE_JPEG ? PIX_FMT_YUVJ422P : PIX_FMT_YUV422P;
                 }else{
                     s->avctx->pix_fmt = s->avctx->get_format(s->avctx,
                                                              s->avctx->codec->pix_fmts ?
@@ -2582,6 +2797,11 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
                                                              hwaccel_pixfmt_list_h264_jpeg_420 :
                                                              ff_hwaccel_pixfmt_list_420);
                 }
+                break;
+            default:
+                av_log(s->avctx, AV_LOG_ERROR,
+                       "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
+                return AVERROR_INVALIDDATA;
         }
 
         s->avctx->hwaccel = ff_find_hwaccel(s->avctx->codec->id, s->avctx->pix_fmt);
@@ -2594,7 +2814,10 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         h->prev_interlaced_frame = 1;
 
         init_scan_tables(h);
-        ff_h264_alloc_tables(h);
+        if (ff_h264_alloc_tables(h) < 0) {
+            av_log(h->s.avctx, AV_LOG_ERROR, "Could not allocate memory for h264\n");
+            return AVERROR(ENOMEM);
+        }
 
         if (!HAVE_THREADS || !(s->avctx->active_thread_type&FF_THREAD_SLICE)) {
             if (context_init(h) < 0) {
@@ -2602,7 +2825,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
                 return -1;
             }
         } else {
-            for(i = 1; i < s->avctx->thread_count; i++) {
+            for(i = 1; i < s->slice_context_count; i++) {
                 H264Context *c;
                 c = h->thread_context[i] = av_malloc(sizeof(H264Context));
                 memcpy(c, h->s.thread_context[i], sizeof(MpegEncContext));
@@ -2615,7 +2838,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
                 clone_tables(c, h, i);
             }
 
-            for(i = 0; i < s->avctx->thread_count; i++)
+            for(i = 0; i < s->slice_context_count; i++)
                 if (context_init(h->thread_context[i]) < 0) {
                     av_log(h->s.avctx, AV_LOG_ERROR, "context_init() failed.\n");
                     return -1;
@@ -2623,6 +2846,11 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         }
     }
 
+    if(h == h0 && h->dequant_coeff_pps != pps_id){
+        h->dequant_coeff_pps = pps_id;
+        init_dequant_tables(h);
+    }
+
     h->frame_num= get_bits(&s->gb, h->sps.log2_max_frame_num);
 
     h->mb_mbaff = 0;
@@ -2668,12 +2896,14 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 0);
             ff_thread_report_progress((AVFrame*)s->current_picture_ptr, INT_MAX, 1);
             ff_generate_sliding_window_mmcos(h);
-            ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
+            if (ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index) < 0 &&
+                (s->avctx->err_recognition & AV_EF_EXPLODE))
+                return AVERROR_INVALIDDATA;
             /* Error concealment: if a ref is missing, copy the previous ref in its place.
              * FIXME: avoiding a memcpy would be nice, but ref handling makes many assumptions
              * about there being no actual duplicates.
              * FIXME: this doesn't copy padding for out-of-frame motion vectors.  Given we're
-             * concealing a lost frame, this probably isn't noticable by comparison, but it should
+             * concealing a lost frame, this probably isn't noticeable by comparison, but it should
              * be fixed. */
             if (h->short_ref_count) {
                 if (prev) {
@@ -2690,7 +2920,7 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         if (s0->first_field) {
             assert(s0->current_picture_ptr);
             assert(s0->current_picture_ptr->f.data[0]);
-            assert(s0->current_picture_ptr->reference != DELAYED_PIC_REF);
+            assert(s0->current_picture_ptr->f.reference != DELAYED_PIC_REF);
 
             /* figure out if we have a complementary field pair */
             if (!FIELD_PICTURE || s->picture_structure == last_pic_structure) {
@@ -2790,6 +3020,8 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
     h->ref_count[1]= h->pps.ref_count[1];
 
     if(h->slice_type_nos != AV_PICTURE_TYPE_I){
+        int max_refs = s->picture_structure == PICT_FRAME ? 16 : 32;
+
         if(h->slice_type_nos == AV_PICTURE_TYPE_B){
             h->direct_spatial_mv_pred= get_bits1(&s->gb);
         }
@@ -2799,13 +3031,14 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
             h->ref_count[0]= get_ue_golomb(&s->gb) + 1;
             if(h->slice_type_nos==AV_PICTURE_TYPE_B)
                 h->ref_count[1]= get_ue_golomb(&s->gb) + 1;
+        }
 
-            if(h->ref_count[0]-1 > 32-1 || h->ref_count[1]-1 > 32-1){
-                av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
-                h->ref_count[0]= h->ref_count[1]= 1;
-                return -1;
-            }
+        if (h->ref_count[0] > max_refs || h->ref_count[1] > max_refs) {
+            av_log(h->s.avctx, AV_LOG_ERROR, "reference overflow\n");
+            h->ref_count[0] = h->ref_count[1] = 1;
+            return AVERROR_INVALIDDATA;
         }
+
         if(h->slice_type_nos == AV_PICTURE_TYPE_B)
             h->list_count= 2;
         else
@@ -2817,8 +3050,10 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         ff_h264_fill_default_ref_list(h);
     }
 
-    if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0)
+    if(h->slice_type_nos!=AV_PICTURE_TYPE_I && ff_h264_decode_ref_pic_list_reordering(h) < 0) {
+        h->ref_count[1]= h->ref_count[0]= 0;
         return -1;
+    }
 
     if(h->slice_type_nos!=AV_PICTURE_TYPE_I){
         s->last_picture_ptr= &h->ref_list[0][0];
@@ -2842,8 +3077,9 @@ static int decode_slice_header(H264Context *h, H264Context *h0){
         }
     }
 
-    if(h->nal_ref_idc)
-        ff_h264_decode_ref_pic_marking(h0, &s->gb);
+    if(h->nal_ref_idc && ff_h264_decode_ref_pic_marking(h0, &s->gb) < 0 &&
+       (s->avctx->err_recognition & AV_EF_EXPLODE))
+        return AVERROR_INVALIDDATA;
 
     if(FRAME_MBAFF){
         ff_h264_fill_mbaff_ref_list(h);
@@ -3093,7 +3329,7 @@ static av_always_inline void fill_filter_caches_inter(H264Context *h, MpegEncCon
 
 /**
  *
- * @return non zero if the loop filter can be skiped
+ * @return non zero if the loop filter can be skipped
  */
 static int fill_filter_caches(H264Context *h, int mb_type){
     MpegEncContext * const s = &h->s;
@@ -3229,6 +3465,7 @@ static void loop_filter(H264Context *h, int start_x, int end_x){
     const int end_mb_y= s->mb_y + FRAME_MBAFF;
     const int old_slice_type= h->slice_type;
     const int pixel_shift = h->pixel_shift;
+    const int block_h = 16 >> s->chroma_y_shift;
 
     if(h->deblocking_filter) {
         for(mb_x= start_x; mb_x<end_x; mb_x++){
@@ -3245,8 +3482,8 @@ static void loop_filter(H264Context *h, int start_x, int end_x){
                 s->mb_x= mb_x;
                 s->mb_y= mb_y;
                 dest_y  = s->current_picture.f.data[0] + ((mb_x << pixel_shift) + mb_y * s->linesize  ) * 16;
-                dest_cb = s->current_picture.f.data[1] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * (8 << CHROMA444);
-                dest_cr = s->current_picture.f.data[2] + ((mb_x << pixel_shift) + mb_y * s->uvlinesize) * (8 << CHROMA444);
+                dest_cb = s->current_picture.f.data[1] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
+                dest_cr = s->current_picture.f.data[2] + (mb_x << pixel_shift) * (8 << CHROMA444) + mb_y * s->uvlinesize * block_h;
                     //FIXME simplify above
 
                 if (MB_FIELD) {
@@ -3254,14 +3491,14 @@ static void loop_filter(H264Context *h, int start_x, int end_x){
                     uvlinesize = h->mb_uvlinesize = s->uvlinesize * 2;
                     if(mb_y&1){ //FIXME move out of this function?
                         dest_y -= s->linesize*15;
-                        dest_cb-= s->uvlinesize*((8 << CHROMA444)-1);
-                        dest_cr-= s->uvlinesize*((8 << CHROMA444)-1);
+                        dest_cb-= s->uvlinesize * (block_h - 1);
+                        dest_cr-= s->uvlinesize * (block_h - 1);
                     }
                 } else {
                     linesize   = h->mb_linesize   = s->linesize;
                     uvlinesize = h->mb_uvlinesize = s->uvlinesize;
                 }
-                backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, CHROMA444, 0);
+                backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize, uvlinesize, 0);
                 if(fill_filter_caches(h, mb_type))
                     continue;
                 h->chroma_qp[0] = get_chroma_qp(h, 0, s->current_picture.f.qscale_table[mb_xy]);
@@ -3330,7 +3567,7 @@ static void decode_finish_row(H264Context *h){
 static int decode_slice(struct AVCodecContext *avctx, void *arg){
     H264Context *h = *(void**)arg;
     MpegEncContext * const s = &h->s;
-    const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
+    const int part_mask= s->partitioned_frame ? (ER_AC_END|ER_AC_ERROR) : 0x7F;
     int lf_x_start = s->mb_x;
 
     s->mb_skip_run= -1;
@@ -3369,13 +3606,13 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             eos = get_cabac_terminate( &h->cabac );
 
             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
-                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
                 if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
                 return 0;
             }
             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d, bytestream (%td)\n", s->mb_x, s->mb_y, h->cabac.bytestream_end - h->cabac.bytestream);
-                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
+                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
                 return -1;
             }
 
@@ -3393,7 +3630,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
 
             if( eos || s->mb_y >= s->mb_height ) {
                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
-                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
                 if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
                 return 0;
             }
@@ -3415,7 +3652,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
 
             if(ret<0){
                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
-                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
+                ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
                 return -1;
             }
 
@@ -3432,27 +3669,27 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
                 if(s->mb_y >= s->mb_height){
                     tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
 
-                    if(get_bits_count(&s->gb) == s->gb.size_in_bits ) {
-                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                    if (get_bits_left(&s->gb) == 0) {
+                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
 
                         return 0;
                     }else{
-                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                        ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_END&part_mask);
 
                         return -1;
                     }
                 }
             }
 
-            if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
+            if (get_bits_left(&s->gb) <= 0 && s->mb_skip_run <= 0){
                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
-                if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
-                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                if (get_bits_left(&s->gb) == 0) {
+                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, ER_MB_END&part_mask);
                     if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
 
                     return 0;
                 }else{
-                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
+                    ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, ER_MB_ERROR&part_mask);
 
                     return -1;
                 }
@@ -3467,26 +3704,24 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
  * @param h h264 master context
  * @param context_count number of contexts to execute
  */
-static void execute_decode_slices(H264Context *h, int context_count){
+static int execute_decode_slices(H264Context *h, int context_count){
     MpegEncContext * const s = &h->s;
     AVCodecContext * const avctx= s->avctx;
     H264Context *hx;
     int i;
 
-    if (s->avctx->hwaccel)
-        return;
-    if(s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
-        return;
+    if (s->avctx->hwaccel || s->avctx->codec->capabilities&CODEC_CAP_HWACCEL_VDPAU)
+        return 0;
     if(context_count == 1) {
-        decode_slice(avctx, &h);
+        return decode_slice(avctx, &h);
     } else {
         for(i = 1; i < context_count; i++) {
             hx = h->thread_context[i];
-            hx->s.error_recognition = avctx->error_recognition;
+            hx->s.err_recognition = avctx->err_recognition;
             hx->s.error_count = 0;
         }
 
-        avctx->execute(avctx, (void *)decode_slice,
+        avctx->execute(avctx, decode_slice,
                        h->thread_context, NULL, context_count, sizeof(void*));
 
         /* pull back stuff from slices to master context */
@@ -3498,6 +3733,8 @@ static void execute_decode_slices(H264Context *h, int context_count){
         for(i = 1; i < context_count; i++)
             h->s.error_count += h->thread_context[i]->s.error_count;
     }
+
+    return 0;
 }
 
 
@@ -3512,7 +3749,7 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
     int nals_needed=0; ///< number of NALs that need decoding before the next frame thread starts
     int nal_index;
 
-    h->max_contexts = (HAVE_THREADS && (s->avctx->active_thread_type&FF_THREAD_SLICE)) ? avctx->thread_count : 1;
+    h->max_contexts = s->slice_context_count;
     if(!(s->flags2 & CODEC_FLAG2_CHUNKS)){
         h->current_slice = 0;
         if (!s->first_field)
@@ -3534,7 +3771,7 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
         int err;
 
         if(buf_index >= next_avc) {
-            if(buf_index >= buf_size) break;
+            if (buf_index >= buf_size - h->nal_length_size) break;
             nalsize = 0;
             for(i = 0; i < h->nal_length_size; i++)
                 nalsize = (nalsize << 8) | buf[buf_index++];
@@ -3592,9 +3829,13 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
             switch (hx->nal_unit_type) {
                 case NAL_SPS:
                 case NAL_PPS:
+                    nals_needed = nal_index;
+                    break;
                 case NAL_IDR_SLICE:
                 case NAL_SLICE:
-                    nals_needed = nal_index;
+                    init_get_bits(&hx->s.gb, ptr, bit_length);
+                    if (!get_ue_golomb(&hx->s.gb))
+                        nals_needed = nal_index;
             }
             continue;
         }
@@ -3611,7 +3852,7 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
                 av_log(h->s.avctx, AV_LOG_ERROR, "Invalid mix of idr and non-idr slices");
                 return -1;
             }
-            idr(h); //FIXME ensure we don't loose some frames if there is reordering
+            idr(h); // FIXME ensure we don't lose some frames if there is reordering
         case NAL_SLICE:
             init_get_bits(&hx->s.gb, ptr, bit_length);
             hx->intra_gb_ptr=
@@ -3686,7 +3927,14 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
             break;
         case NAL_SPS:
             init_get_bits(&s->gb, ptr, bit_length);
-            ff_h264_decode_seq_parameter_set(h);
+            if (ff_h264_decode_seq_parameter_set(h) < 0 &&
+                h->is_avc && (nalsize != consumed) && nalsize) {
+                av_log(h->s.avctx, AV_LOG_DEBUG, "SPS decoding failure, "
+                       "try parsing the coomplete NAL\n");
+                init_get_bits(&s->gb, buf + buf_index + 1 - consumed,
+                              8 * (nalsize - 1));
+                ff_h264_decode_seq_parameter_set(h);
+            }
 
             if (s->flags& CODEC_FLAG_LOW_DELAY ||
                 (h->sps.bitstream_restriction_flag && !h->sps.num_reorder_frames))
@@ -3695,17 +3943,19 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
             if(avctx->has_b_frames < 2)
                 avctx->has_b_frames= !s->low_delay;
 
-            if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma) {
+            if (avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
+                h->cur_chroma_format_idc != h->sps.chroma_format_idc) {
                 if (h->sps.bit_depth_luma >= 8 && h->sps.bit_depth_luma <= 10) {
                     avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
+                    h->cur_chroma_format_idc = h->sps.chroma_format_idc;
                     h->pixel_shift = h->sps.bit_depth_luma > 8;
 
-                    ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma);
-                    ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma);
+                    ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
+                    ff_h264_pred_init(&h->hpc, s->codec_id, h->sps.bit_depth_luma, h->sps.chroma_format_idc);
                     s->dsp.dct_bits = h->sps.bit_depth_luma > 8 ? 32 : 16;
                     dsputil_init(&s->dsp, s->avctx);
                 } else {
-                    av_log(avctx, AV_LOG_DEBUG, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
+                    av_log(avctx, AV_LOG_ERROR, "Unsupported bit depth: %d\n", h->sps.bit_depth_luma);
                     return -1;
                 }
             }
@@ -3752,7 +4002,7 @@ static int decode_nal_units(H264Context *h, const uint8_t *buf, int buf_size){
 }
 
 /**
- * returns the number of bytes consumed for building the current frame
+ * Return the number of bytes consumed for building the current frame.
  */
 static int get_consumed_bytes(MpegEncContext *s, int pos, int buf_size){
         if(pos==0) pos=1; //avoid infinite loops (i doubt that is needed but ...)
@@ -3770,7 +4020,7 @@ static int decode_frame(AVCodecContext *avctx,
     H264Context *h = avctx->priv_data;
     MpegEncContext *s = &h->s;
     AVFrame *pict = data;
-    int buf_index;
+    int buf_index = 0;
 
     s->flags= avctx->flags;
     s->flags2= avctx->flags2;
@@ -3800,7 +4050,7 @@ static int decode_frame(AVCodecContext *avctx,
             *pict= *(AVFrame*)out;
         }
 
-        return 0;
+        return buf_index;
     }
 
     buf_index=decode_nal_units(h, buf, buf_size);
@@ -3871,10 +4121,10 @@ int main(void){
     uint8_t temp[SIZE];
     PutBitContext pb;
     GetBitContext gb;
-//    int int_temp[10000];
     DSPContext dsp;
     AVCodecContext avctx;
 
+    avctx.av_class = avcodec_get_class();
     dsputil_init(&dsp, &avctx);
 
     init_put_bits(&pb, temp, SIZE);
@@ -3888,9 +4138,7 @@ int main(void){
 
     init_get_bits(&gb, temp, 8*SIZE);
     for(i=0; i<COUNT; i++){
-        int j, s;
-
-        s= show_bits(&gb, 24);
+        int j, s = show_bits(&gb, 24);
 
         START_TIMER
         j= get_ue_golomb(&gb);
@@ -3913,9 +4161,7 @@ int main(void){
 
     init_get_bits(&gb, temp, 8*SIZE);
     for(i=0; i<COUNT; i++){
-        int j, s;
-
-        s= show_bits(&gb, 24);
+        int j, s = show_bits(&gb, 24);
 
         START_TIMER
         j= get_se_golomb(&gb);
@@ -3979,16 +4225,15 @@ static const AVProfile profiles[] = {
 };
 
 AVCodec ff_h264_decoder = {
-    "h264",
-    AVMEDIA_TYPE_VIDEO,
-    CODEC_ID_H264,
-    sizeof(H264Context),
-    ff_h264_decode_init,
-    NULL,
-    ff_h264_decode_end,
-    decode_frame,
-    /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
-        CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
+    .name           = "h264",
+    .type           = AVMEDIA_TYPE_VIDEO,
+    .id             = CODEC_ID_H264,
+    .priv_data_size = sizeof(H264Context),
+    .init           = ff_h264_decode_init,
+    .close          = ff_h264_decode_end,
+    .decode         = decode_frame,
+    .capabilities   = /*CODEC_CAP_DRAW_HORIZ_BAND |*/ CODEC_CAP_DR1 | CODEC_CAP_DELAY |
+                      CODEC_CAP_SLICE_THREADS | CODEC_CAP_FRAME_THREADS,
     .flush= flush_dpb,
     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10"),
     .init_thread_copy      = ONLY_IF_THREADS_ENABLED(decode_init_thread_copy),
@@ -3998,15 +4243,14 @@ AVCodec ff_h264_decoder = {
 
 #if CONFIG_H264_VDPAU_DECODER
 AVCodec ff_h264_vdpau_decoder = {
-    "h264_vdpau",
-    AVMEDIA_TYPE_VIDEO,
-    CODEC_ID_H264,
-    sizeof(H264Context),
-    ff_h264_decode_init,
-    NULL,
-    ff_h264_decode_end,
-    decode_frame,
-    CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
+    .name           = "h264_vdpau",
+    .type           = AVMEDIA_TYPE_VIDEO,
+    .id             = CODEC_ID_H264,
+    .priv_data_size = sizeof(H264Context),
+    .init           = ff_h264_decode_init,
+    .close          = ff_h264_decode_end,
+    .decode         = decode_frame,
+    .capabilities   = CODEC_CAP_DR1 | CODEC_CAP_DELAY | CODEC_CAP_HWACCEL_VDPAU,
     .flush= flush_dpb,
     .long_name = NULL_IF_CONFIG_SMALL("H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10 (VDPAU acceleration)"),
     .pix_fmts = (const enum PixelFormat[]){PIX_FMT_VDPAU_H264, PIX_FMT_NONE},